A month ago, I was staging the Interop Labs, with a particular personal focus on VoIP on 802.11. At that point, I promised to investigate what improvement Wi-Fi Multimedia (WMM) prioritization makes to voice quality. The Interop show floor is a terrible radio environment. From the Interop Labs booth, I can see between 40 and 70 access points. In this environment, WMM makes a huge difference.
We’re running the Interop Labs on Asterisk. To show off the difference in voice quality, I configured Asterisk to record the incoming voice stream from the 802.11 phone. You can listen to the results, and see what a difference WMM makes. Both recordings were made just after lunch (about 1:20 pm) on the second day of the show, which was one of the busiest times for us in the booth.
So, without further ado, here are the three recordings:
- Introductory remarks. Before the show opened, I recorded a 5-minute introduction to the technology and the challenges we hoped to investigate. This recording was made from a wired phone about ten minutes before the expo floor opened.
- Non-prioritized voice. This recording is a bit over one minute long, and was made between a phone and an access point that don’t support WMM. You can hear the effects of the heavy background traffic in the frequent dropouts.
- WMM-prioritized voice. In the final 2-minute long recording, the voice is prioritized with WMM between the Unex WP2 phone and Trapeze wireless infrastructure. (This is one of the first, if not the first, public interoperable WMM demonstration.) Although the conversation breaks up, it happens less frequently and the dropouts are far less severe.
One important note: I recorded these samples at the PBX, so there’s only one wireless hop in them. A call between two wireless phones would have double the opportunity for packet loss or delay, and the difference between prioritized an unprioritized traffic would be even more striking.
Hi Matthew,
Thanks for the posting. I listened to all the samples. Clearly the one without prioritization suffers in quality. But I have couple of things to mention here. Firstly when you claim 40 to 70 APs in the show ground, I guess only a few will be interfering (WLAN-WLAN interference) with a particular AP at a time because of same channel being used. So even if the traffic is high on the overall network, the traffic in the few interfering AP's maybe less. It would have been nice to see how many APs in the vicinity were interfering with the AP you were using.
Secondly, was the question of mobility addressed ?. Obviously the main reason to switch to 802.11 based WMM is that the option of mobility is available. Did you move around with the 802.11e phone during the call ?. Going out of range from the AP and walking into the point of interference between neighbouring AP's may seriously affect the voice quality. Did you verify any mobility at all ?.
Your experiments are really helpful for students.
Thanks
Arvind
arvindsc@yahoo.com
Arvind,
Of the 40-70 APs that I saw, maybe half were co-channel. I deliberately put the test AP on the busiest channel, and channel 6 is the most common default channel setting.
The important statistic is not the number of interfering APs, but the percentage of time that the shared radio channel is used for transmission. I worked with a couple of different tools to measure channel load, but neither was available at the precise moment I made the recordings. From other observations, channel utilization was consistently above 50%, often bursting much higher.
I did not move far from the AP serving me. To prevent other people from demonstrating the mobility of handheld 802.11 phones to us (in a process commonly labeled "theft"), we had them tied down. I disagree, however, that the main reason to use WMM is mobility. The reason to use WMM is that it improves quality, regardless of mobility. Given the choice between the two calls I posted, you'll take the quality of the second one.
Hi Matthew,
Let me start of by saying that your post is very interesting. I have a couple of questions and would appreciate it if you could answer them.
Firstly, WMM uses EDCA and not HCCA, is that correct? In any case, do you know what were the contention parameter values used for the data packets and that used for the voice packets.
Secondly, were any tests done in which voice calls were made from different phones (i.e, all the calls are assigned the highest priority)? Also, was the effect on the data packets being sent on the network (throughput starvation for instance) analyzed?
Look forward to your reply.
Arvind
swamina@clemson.edu
Matthew: This is (and I hate to use this term, but it really fits) awesome data. People have been saying for years that QoS and VoIP must go hand-in-hand, and this is clear evidence of how important that is. I think that a lot of folks have secretly hoped that we'd be able to over-engineer our way out of QoS for VoIP, but the reports, experiences, and tests like yours keep coming in and showing that it's just not going to be that easy. We have our work cut out for us.
Next steps, I think, would be to compare WMM with other QoS schemes, although the evidence you've got here at least empirically suggests that this is a front-runner.
Thanks again for the data!
Hi Matthew,
You mentioned WMM, and you mentioned that there was background traffic. Could you mention, however, what percentage of the background traffic was WMM? It is well known that WMM works well when there is a limited number of priority-tagged flows, when the background traffic is all best-effort; however, WMM fails when the background traffic is also prioritized, as you would have in a reasonably-dense voice network.